Dear All,
I am trrying to estimate a binary model logit/probit and examine
those factors mostly affecting the binary outcome. The bulk of
independent variables are dummies as they have been generated
by categorical variables. Age is the exemption which is used
either as a continuous or a series of dummies.
Provided that the parameter estimates will be used to construct a
predictor, I need to test the predictive power of the model. For this
reason, double cross validation method is used. Because of this, I
split the sample randomly into two halves and apply the method. In
so doing, I am using the stepwise procedure with logit and
backward selection to each estimation sample (i.e., to each halve)
and to the whole data-set. That is, I use the top-to-bottom selection
method using STATA viz:
sw logistic y x if group==1 (or 2), pr (.10)
For each estimation sample, STATA has to drop one or another
variable due to estimability problems such as 'X failures and Y
successes completely determined' and fairy enough so far. The
problem which arise is how would I conciliate the slope coefficient
for each estimation stage. To be more specific, for the 1st sample,
X1 or X2 are dropped from the estimation stage due to estimability
problems. Conversely, these two variables may well appear on the
final results produced for the 2nd sample. Likewise, variables X3
and X4 are dropped from the 2nd sample at the estimation stage
but both appear at the estimation done for the whole data-set.
Due to the above, I do not really know how to overcome this
obstacle. I realise that this is a data problem and so it has nothing
to do with the capabilities of STATA. Of course, you could alter
the seed for splitting randomly the sample but to me, this does not
sound pretty promising. Accordingly, I have no clue on how to proceed.
I would thus be greatful if you have any ideas and suggestions on
this issue. I would very much value your comments and advice.
Thank you very much indeed for taking the time to consider my
request. I look forward to hearing from you.
Yours sincerely
Panos.
------- End of forwarded message -------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|